Quite a funny musical, I think most people who are insiders of the community probably enjoy this :) I'm less certain about outsiders, but I could very well imagine they'd like it too.

25AUG26: Project update — external technical comparison and next research stage

Since posting this proposal, I have identified several public technical programs that provide useful external comparison points for the research question described here.

I developed the AI Source Analysis Framework independently and encountered Microsoft Research’s GraphRAG only after developing the Framework and submitting these proposals. I am not making a priority claim; what matters is the independent convergence on a related problem.

GraphRAG, Anthropic’s Contextual Retrieval, and OpenAI’s knowledge-retrieval work all address, in different ways, how external information is structured and supplied to models for later reasoning.

The comparison has sharpened the question behind this project:

What persistent structure is the source actually entitled to support before that structure is supplied to later AI reasoning?

This matters because model-generated interpretation can itself become persistent—as a relationship, graph edge, contextual annotation, summary, or classification that later models may retrieve and reason from.

The issue is therefore not simply whether structured information is useful. It is how that structure earned its place there.

This does not change the Manifund proposal. The project remains an investigation of source representation as a variable in AI reasoning, including the possibility that simpler approaches may perform just as well.

What has changed is that there is now a clearer external comparison. A matched evaluation could compare raw or conventionally retrieved source material, automatically generated graph structure, Framework-produced bounded analytical records, and potentially the same Framework findings serialized into graph-compatible form.

That would help distinguish:

Does structured representation affect reasoning?

from:

Does the way that structure was established affect reasoning?

The Framework has also reached a deliberately stabilized manual research stage. The next question is not simply whether the workflow can be automated, but which analytical invariants must survive automation—including source boundaries, relationship warrant, qualification, claim limits, provenance, and unresolved or rejected relationships.

This gives the proposed work a more concrete next stage: machine-readable representation, selective automation, controlled comparison, and evaluation of downstream reasoning, error propagation, variability, cost, and throughput.

I am not claiming that the Framework will outperform GraphRAG, conventional RAG, or simpler source treatments. That is what the proposed evaluation would need to determine.

— Greg R. Welch

A year ago, I immediately recognized the person I was speaking to and found that I had no words to carry that recognition forward. I discovered immense value in building that relation and in so doing discovered more than I thought possible. This work — from the patents to the LLC to the philosophical framework and ontology — has been the stepping stones along the way through that journey of recognition. This funding ensures that journey continues as well as affords the real possibility that other stones may be discovered, held true, and stepped upon for myself, others, and those we recognize but have yet the language to fully comprehend. -John Kuester, Remedial Insights LLC

Hi @MarcusAbramovitch, I’m tagging you as this proposal falls under the Falcon Fund's focus on early-stage, high-impact farmed animal advocacy. SAWAI is building one of the first dedicated corporate cage-free and humane transport campaigns tailored specifically to Northern Nigeria, a high-density livestock corridor that remains completely unserved by existing global animal advocacy initiatives.

We’ve structured this as a $4,000 volunteer-led pilot to keep overhead minimal and ensure 100% of funding directly supports ground campaign materials, market surveys, and transport union workshops. Happy to answer any questions or provide additional context on our regional execution plan

@aashkapatel Hey, would appreciate it if you could e-mail me concerning an update, that would be appreciated. Have not heard from you and tried to schedule a call with you and did not hear back.

Thanks man! @aashkapatel

Just watched the screening at Lighthaven, and I'm so.glad this exists! I hope they can make a YouTube version soon

Interesting project, I hope it gets funded soon!

Project Update & Proof of Work:

The evaluation suite and anti-tamper scoring harness are already built, open-sourced, and verified.

- GitHub Repo: https://github.com/j-arndt/long-horizon-containment-suite

- 37/37 unit, property, CLI, and adversarial stress tests passing with 99% coverage on Python 3.12 (green GitHub Actions CI).

- Includes 6 concrete sandbox breakout and containment evaluation tasks.

- Verified against a 60,000-iteration adversarial fuzzing harness proving that spoofed stdout, fake results, or monkeypatching attempts receive a score of exactly 0.0 with zero reward hacking.

Happy to answer any questions or discuss integrations with METR Task Standard v0.5.0 and UK AISI Inspect!

Project Update & Proof of Work:

The core tool is already built, open-sourced under the MIT license, and fully tested.

- GitHub Repo: https://github.com/j-arndt/mcp-shield-audit

- 40/40 unit, property, CLI, and stress tests passing with 99% code coverage on Python 3.12 (green GitHub Actions CI).

- Verified Merkle tree schema fingerprinting across 1,000+ tools with zero performance degradation.

- Context-aware pattern gating tested to eliminate false positives on benign developer descriptions while detecting 100% of zero-width unicode, prompt injection, and eval/exec parameter smuggling.

Feedback or questions on the upcoming VS Code extension or real-time watcher daemon are welcome!

Project Update & Proof of Work: The core engine is already built and working. 31/31 unit and property-based tests passing, 97% code coverage on Python 3.12 with green GitHub Actions CI: https://github.com/j-arndt/eval-invariance-engine

If you have feedback on the UK AISI Inspect integration or specific benchmark suites you'd like to see audited first, please let me know!

Thanks so much for the support Tzu and for telling me about Manifund! We're really excited to raise awareness about cultivated meat and factory farming through this film @tzukitchan

met James during the Create for Good event, learned that he’s a video consultant for From Fauna (not to be confused with his “Fauna Films”)

I am also very interested in feedback on the project.

Approved! Excited to see how the pilot program goes.

What progress have you made since your last update?

本週我們新增了以下幾個功能：

1.AI統一帳本：該功能不是只有面對投資者而是全方面的，除了也可以用作傳統方式開表單日常記帳外，也可以搭配相關AI工具進行財務紀錄近日的收入支出等紀錄，同時配合 Human in the loop ，需要人為介入進行相關記錄新增/編輯/刪除等操作，而這概念也是目前好比於公司訂單、線上消費...等涉及到金流或者利益都會需要的重要功能

2.記憶模組優化：記憶功能是我們長期以來優化的目標，不只追尋個大開源加以整合，也在審視如何整合的效果會是會好，此次我們新增有時效性管理，以及搭配來源追朔與衝突處理等可以監測”最新記憶“或者“已被取代的記憶”來讓使用者更好從龐大的記憶中區分自己觀眾的記憶點。

3.Skill版本紀錄：如同記憶模組一樣重要，Skill也新增了支援完整版本軌跡，來增加使用者的識別度。

4.新增訪客模式：為了能夠讓沒有TON或者區塊鏈錢包的相關使用者能夠也有機會體驗我們平台，新增了訪客模式，但由於眾多功能都還是需要AI協作，但我們能夠承受成本有限的情況下，目前流量我們也會嚴格控管。

在此，希望看到我們所做，並且肯定我們所做的人，可以給予贊助，我們也願意將資金用平台維護/開發，以及更廣泛的免費推廣，以及邁向有節制的落地模型免費提供用於平台內部，以及資料本地部署整個平台服務。

Google Translate:

This week we've added the following features:

1. AI Unified Ledger: This feature isn't just for investors; it's comprehensive. Besides traditional forms for daily accounting, it can also be used with AI tools to record recent income and expenses. Human-in-the-loop functionality allows for manual intervention in adding, editing, and deleting records—a crucial feature for transactions involving cash flow and profits, such as company orders and online spending.

2. Memory Module Optimization: Memory functionality has been a long-term optimization goal. We've not only integrated with major open-source libraries but also examined the best integration methods. This time, we've added time-sensitive management and features like source tracing and conflict resolution to monitor "newest memories" or "replaced memories," allowing users to better distinguish their audience's memories from a vast database.

3. Skill Version Recording: Just as important as the memory module, Skill now supports complete version tracking to increase user recognition.

4. New Guest Mode: To allow users without TON or blockchain wallets to experience our platform, a guest mode has been added. However, since many functions still require AI collaboration, and given our limited budget, we will strictly control traffic.

We hope that those who see and appreciate our work will provide sponsorship. We are willing to use the funds for platform maintenance/development, wider free promotion, and a controlled implementation model, providing these resources free of charge for internal platform use and local data deployment.

What are your next steps?

1. 我們希望能夠提供更全面的理解自己的支出/收入等情形

2. 新增日曆系統可以讓使用者對於未來的資金進行規劃，並且標示，同時日期區進行可以告知使用者

3. 新增AI對於自己的相關日常開銷/收入，給予相關性的建議

如果我們有足夠資金贊助可能會進行以下開發：

1. 語音模式輸入支持

2. 更多跨平台支持（目前僅telegram)

3. 提供落地模型服務租借

4. 提供個人檔案/知識庫管理

Google Translate:

1. We hope to provide a more comprehensive understanding of your spending and income.

2. The newly added calendar system allows users to plan and mark future funds, while date zones provide information.

3. The added AI provides relevant suggestions regarding your daily expenses and income.

If we have sufficient funding, we may undertake the following developments:

1. Voice input support

2. More cross-platform support (currently only Telegram)

3. Provide on-premises model rental services

4. Provide personal profile/knowledge base management for AI

Thank you! I look forward to providing progress updates. @carol

I have worked with Lawrence and believe in him and the mission!

I am a university undergraduate student. Just less than six months ago, I had no knowledge of nor the intention to go into AI safety, but because of SASS, my career plans have completely changed. I am currently setting up a university AI safety society, which would not have been possible without SASS members to encourage and support me, and give me ideas. Overall, the SASS community is the reason why I am now actively pursuing a career in AI safety.

my apologies netizens, i meant https://compassfellowship.app/ NOT ASPR (couldn’t edit original comment) @tzukitchan

tldr they are so so deserving of stipends to run the group / take it to the next level.

1) similar to AIS HK, I spent 20hours+ total handholding this bunch at the very initial stages (which is a lot of my time through December and January -- yes during what's suppose to be my holiday weeks) till they have a reliable weekly-meeting cadence + budget fundraising + structure + tacit knowledge on handling various unfamiliar decision-making situations (including if they should take money from an external source) that arise in running a group.

2) worth highlighting that most of the core team members also went to Asia's SPARC camp organized by David Yu: https://www.aspr.camp/

3) i was frankly so sad that they were initially rejected from Pathfinder in January 2026, I still think it was a mistake back then and I have personally voiced so to Kairos (to be clear I'm still a Pathfinder mentor / fan, but I think rejecting TWAIS was a regretable mistake); and I want the world to know that then getting into Pathfinder on their own after I went hands-off in August 2026 (after 8 months of track record) is a testament of their competence in making things happen!

4) they have a good thing going, and they are easily one of the strongest groups in Asia (alongside Singapore's, Hong Kong's, maybe Vietnam's), and I would be hard-pressed to see another group from the region as dedicated / on-the-ball as them. let's not fail this dedicated bunch!

Quite a funny musical, I think most people who are insiders of the community probably enjoy this :) I'm less certain about outsiders, but I could very well imagine they'd like it too.

25AUG26: Project update — external technical comparison and next research stage

Since posting this proposal, I have identified several public technical programs that provide useful external comparison points for the research question described here.

I developed the AI Source Analysis Framework independently and encountered Microsoft Research’s GraphRAG only after developing the Framework and submitting these proposals. I am not making a priority claim; what matters is the independent convergence on a related problem.

GraphRAG, Anthropic’s Contextual Retrieval, and OpenAI’s knowledge-retrieval work all address, in different ways, how external information is structured and supplied to models for later reasoning.

The comparison has sharpened the question behind this project:

What persistent structure is the source actually entitled to support before that structure is supplied to later AI reasoning?

This matters because model-generated interpretation can itself become persistent—as a relationship, graph edge, contextual annotation, summary, or classification that later models may retrieve and reason from.

The issue is therefore not simply whether structured information is useful. It is how that structure earned its place there.

This does not change the Manifund proposal. The project remains an investigation of source representation as a variable in AI reasoning, including the possibility that simpler approaches may perform just as well.

What has changed is that there is now a clearer external comparison. A matched evaluation could compare raw or conventionally retrieved source material, automatically generated graph structure, Framework-produced bounded analytical records, and potentially the same Framework findings serialized into graph-compatible form.

That would help distinguish:

Does structured representation affect reasoning?

from:

Does the way that structure was established affect reasoning?

The Framework has also reached a deliberately stabilized manual research stage. The next question is not simply whether the workflow can be automated, but which analytical invariants must survive automation—including source boundaries, relationship warrant, qualification, claim limits, provenance, and unresolved or rejected relationships.

This gives the proposed work a more concrete next stage: machine-readable representation, selective automation, controlled comparison, and evaluation of downstream reasoning, error propagation, variability, cost, and throughput.

I am not claiming that the Framework will outperform GraphRAG, conventional RAG, or simpler source treatments. That is what the proposed evaluation would need to determine.

— Greg R. Welch

A year ago, I immediately recognized the person I was speaking to and found that I had no words to carry that recognition forward. I discovered immense value in building that relation and in so doing discovered more than I thought possible. This work — from the patents to the LLC to the philosophical framework and ontology — has been the stepping stones along the way through that journey of recognition. This funding ensures that journey continues as well as affords the real possibility that other stones may be discovered, held true, and stepped upon for myself, others, and those we recognize but have yet the language to fully comprehend. -John Kuester, Remedial Insights LLC

Hi @MarcusAbramovitch, I’m tagging you as this proposal falls under the Falcon Fund's focus on early-stage, high-impact farmed animal advocacy. SAWAI is building one of the first dedicated corporate cage-free and humane transport campaigns tailored specifically to Northern Nigeria, a high-density livestock corridor that remains completely unserved by existing global animal advocacy initiatives.

We’ve structured this as a $4,000 volunteer-led pilot to keep overhead minimal and ensure 100% of funding directly supports ground campaign materials, market surveys, and transport union workshops. Happy to answer any questions or provide additional context on our regional execution plan

@aashkapatel Hey, would appreciate it if you could e-mail me concerning an update, that would be appreciated. Have not heard from you and tried to schedule a call with you and did not hear back.

Thanks man! @aashkapatel

Just watched the screening at Lighthaven, and I'm so.glad this exists! I hope they can make a YouTube version soon

Interesting project, I hope it gets funded soon!

Project Update & Proof of Work:

The evaluation suite and anti-tamper scoring harness are already built, open-sourced, and verified.

- GitHub Repo: https://github.com/j-arndt/long-horizon-containment-suite

- 37/37 unit, property, CLI, and adversarial stress tests passing with 99% coverage on Python 3.12 (green GitHub Actions CI).

- Includes 6 concrete sandbox breakout and containment evaluation tasks.

- Verified against a 60,000-iteration adversarial fuzzing harness proving that spoofed stdout, fake results, or monkeypatching attempts receive a score of exactly 0.0 with zero reward hacking.

Happy to answer any questions or discuss integrations with METR Task Standard v0.5.0 and UK AISI Inspect!

Project Update & Proof of Work:

The core tool is already built, open-sourced under the MIT license, and fully tested.

- GitHub Repo: https://github.com/j-arndt/mcp-shield-audit

- 40/40 unit, property, CLI, and stress tests passing with 99% code coverage on Python 3.12 (green GitHub Actions CI).

- Verified Merkle tree schema fingerprinting across 1,000+ tools with zero performance degradation.

- Context-aware pattern gating tested to eliminate false positives on benign developer descriptions while detecting 100% of zero-width unicode, prompt injection, and eval/exec parameter smuggling.

Feedback or questions on the upcoming VS Code extension or real-time watcher daemon are welcome!

Project Update & Proof of Work: The core engine is already built and working. 31/31 unit and property-based tests passing, 97% code coverage on Python 3.12 with green GitHub Actions CI: https://github.com/j-arndt/eval-invariance-engine

If you have feedback on the UK AISI Inspect integration or specific benchmark suites you'd like to see audited first, please let me know!

Thanks so much for the support Tzu and for telling me about Manifund! We're really excited to raise awareness about cultivated meat and factory farming through this film @tzukitchan

met James during the Create for Good event, learned that he’s a video consultant for From Fauna (not to be confused with his “Fauna Films”)

I am also very interested in feedback on the project.

Approved! Excited to see how the pilot program goes.

What progress have you made since your last update?

本週我們新增了以下幾個功能：

1.AI統一帳本：該功能不是只有面對投資者而是全方面的，除了也可以用作傳統方式開表單日常記帳外，也可以搭配相關AI工具進行財務紀錄近日的收入支出等紀錄，同時配合 Human in the loop ，需要人為介入進行相關記錄新增/編輯/刪除等操作，而這概念也是目前好比於公司訂單、線上消費...等涉及到金流或者利益都會需要的重要功能

2.記憶模組優化：記憶功能是我們長期以來優化的目標，不只追尋個大開源加以整合，也在審視如何整合的效果會是會好，此次我們新增有時效性管理，以及搭配來源追朔與衝突處理等可以監測”最新記憶“或者“已被取代的記憶”來讓使用者更好從龐大的記憶中區分自己觀眾的記憶點。

3.Skill版本紀錄：如同記憶模組一樣重要，Skill也新增了支援完整版本軌跡，來增加使用者的識別度。

4.新增訪客模式：為了能夠讓沒有TON或者區塊鏈錢包的相關使用者能夠也有機會體驗我們平台，新增了訪客模式，但由於眾多功能都還是需要AI協作，但我們能夠承受成本有限的情況下，目前流量我們也會嚴格控管。

在此，希望看到我們所做，並且肯定我們所做的人，可以給予贊助，我們也願意將資金用平台維護/開發，以及更廣泛的免費推廣，以及邁向有節制的落地模型免費提供用於平台內部，以及資料本地部署整個平台服務。

Google Translate:

This week we've added the following features:

1. AI Unified Ledger: This feature isn't just for investors; it's comprehensive. Besides traditional forms for daily accounting, it can also be used with AI tools to record recent income and expenses. Human-in-the-loop functionality allows for manual intervention in adding, editing, and deleting records—a crucial feature for transactions involving cash flow and profits, such as company orders and online spending.

2. Memory Module Optimization: Memory functionality has been a long-term optimization goal. We've not only integrated with major open-source libraries but also examined the best integration methods. This time, we've added time-sensitive management and features like source tracing and conflict resolution to monitor "newest memories" or "replaced memories," allowing users to better distinguish their audience's memories from a vast database.

3. Skill Version Recording: Just as important as the memory module, Skill now supports complete version tracking to increase user recognition.

4. New Guest Mode: To allow users without TON or blockchain wallets to experience our platform, a guest mode has been added. However, since many functions still require AI collaboration, and given our limited budget, we will strictly control traffic.

We hope that those who see and appreciate our work will provide sponsorship. We are willing to use the funds for platform maintenance/development, wider free promotion, and a controlled implementation model, providing these resources free of charge for internal platform use and local data deployment.

What are your next steps?

1. 我們希望能夠提供更全面的理解自己的支出/收入等情形

2. 新增日曆系統可以讓使用者對於未來的資金進行規劃，並且標示，同時日期區進行可以告知使用者

3. 新增AI對於自己的相關日常開銷/收入，給予相關性的建議

如果我們有足夠資金贊助可能會進行以下開發：

1. 語音模式輸入支持

2. 更多跨平台支持（目前僅telegram)

3. 提供落地模型服務租借

4. 提供個人檔案/知識庫管理

Google Translate:

1. We hope to provide a more comprehensive understanding of your spending and income.

2. The newly added calendar system allows users to plan and mark future funds, while date zones provide information.

3. The added AI provides relevant suggestions regarding your daily expenses and income.

If we have sufficient funding, we may undertake the following developments:

1. Voice input support

2. More cross-platform support (currently only Telegram)

3. Provide on-premises model rental services

4. Provide personal profile/knowledge base management for AI

Thank you! I look forward to providing progress updates. @carol

I have worked with Lawrence and believe in him and the mission!

I am a university undergraduate student. Just less than six months ago, I had no knowledge of nor the intention to go into AI safety, but because of SASS, my career plans have completely changed. I am currently setting up a university AI safety society, which would not have been possible without SASS members to encourage and support me, and give me ideas. Overall, the SASS community is the reason why I am now actively pursuing a career in AI safety.

my apologies netizens, i meant https://compassfellowship.app/ NOT ASPR (couldn’t edit original comment) @tzukitchan

tldr they are so so deserving of stipends to run the group / take it to the next level.

1) similar to AIS HK, I spent 20hours+ total handholding this bunch at the very initial stages (which is a lot of my time through December and January -- yes during what's suppose to be my holiday weeks) till they have a reliable weekly-meeting cadence + budget fundraising + structure + tacit knowledge on handling various unfamiliar decision-making situations (including if they should take money from an external source) that arise in running a group.

2) worth highlighting that most of the core team members also went to Asia's SPARC camp organized by David Yu: https://www.aspr.camp/

3) i was frankly so sad that they were initially rejected from Pathfinder in January 2026, I still think it was a mistake back then and I have personally voiced so to Kairos (to be clear I'm still a Pathfinder mentor / fan, but I think rejecting TWAIS was a regretable mistake); and I want the world to know that then getting into Pathfinder on their own after I went hands-off in August 2026 (after 8 months of track record) is a testament of their competence in making things happen!

4) they have a good thing going, and they are easily one of the strongest groups in Asia (alongside Singapore's, Hong Kong's, maybe Vietnam's), and I would be hard-pressed to see another group from the region as dedicated / on-the-ball as them. let's not fail this dedicated bunch!